Study design

After DBS surgery, the electrode leads were externalized to enable recording of LFPs before the electrodes were connected to the IPG a few days later. The experimental protocol included three sessions (one per day) that took place from day 5 through day 7, after the electrode implantation surgery. During the first session, experimenters collected LFP recordings to identify beta band frequencies and to adjust the adaptive deep brain stimulation (aDBS) algorithm accordingly for each patient (see “parametrization”). During the second and the third sessions, we administered the two DBS treatments to patients through the external dual prototype (one treatment per day).The stimulation lasted at least two hours.

Patients were randomly assigned to first receive either cDBS or aDBS. We used computer-generated randomization assignment according to the order of recruitment so that a comparable numbers of patients were treated with aDBS and cDBS. The randomization sequence was generated by the experimenter (the same person as for the aDBS setup). The patient and the neurologist directly involved in the scoring were blinded to the DBS type; only the experimenter was aware of the type of stimulation.

35 3.3.1.3 Experimental procedure

The external prototype was placed in a pouch worn by the patient at the waist level, which allowed the patient to move freely. The device was connected to the electrode selected for the best recording and stimulation.

Both aDBS and cDBS were delivered unilaterally in the hemisphere in which the recorded LFPs showed the highest beta band peak. Accordingly, we recorded the feedback signal for the aDBS from the electrode contact pair that exhibited the highest beta peak and used the contact located between these contacts for stimulation (the same for aDBS and cDBS sessions). At the time of the experiment, levodopa was the only pharmacological treatment administered to the patients who arrived at the surgery after three to four weeks of washout from all their antiparkinsonian medication, according to the current clinical procedure.

Figure 3.2 Experimental sessions

After beta band identification, we selected the best recording and stimulation configuration for optimal aDBS. We determined the hemisphere and electrode contact pairs (0-2; 0-3; 1-3) that exhibited the highest beta peak. If a patient manifested a similar beta peak from both STN signals, we chose the side contralateral to that associated with disease onset in that patient. The stimulation contact was located between the two contacts selected for recording. We then identified the

effective stimulation amplitude for each patient. To this end, we delivered a monopolar stimulation

36

using the prototype in cDBS mode at 130 Hz with a 60 μs pulse and progressively increased the intensity by 0.5 V every 3–4 min until the first stimulation-induced adverse effects appeared. The effective stimulation amplitude was defined as the lowest stimulation intensity that induced at least a 60% improvement in upper limb rigidity on the side contralateral to the stimulation site, without adverse effects.

Then, to establish the reference beta band power values 𝑃_{𝛽𝑂𝐹𝐹} and 𝑃_𝛽², we connected the prototype through the input and output channels to the previously selected electrodes and programmed it to do as follows: record LFPs for about 20 min (baseline clinical condition) before switching on the stimulation at the effective amplitude, to record LFPs for a further 20 min (stimON/medOFF condition) and to continue to record LFPs while administering levodopa (the patient’s usual

antiparkinsonian medication with the treatment dose increased by 50%) and waiting for the drug to achieve its clinical effect (stimON/medON). We then analyzed LFP beta activity offline to extract the reference values of beta band power in the stimOFF/medOFF and stimON/medON,

respectively, 𝑃_{𝛽𝑂𝐹𝐹} and 𝑃_𝛽². Hereafter we refer to the 𝑃_𝛽² value as 𝑃_𝛽𝑂𝑁.

We used the specific beta band boundaries, the effective stimulation amplitude, and the reference values of beta band power for each patient to set up a personalized aDBS algorithm. Because our aim was to reduce the stimulation voltage to a value close to zero, we set 𝑉₂ = 0. Under these conditions, and by substituting in (3) the terms “𝑃_{𝛽𝑂𝐹𝐹} − 𝑃_𝛽” with “𝑃_{𝛽𝑂𝐹𝐹} − 𝑃_𝛽𝑂𝑁− 𝑃_𝛽 + 𝑃_𝛽𝑂𝑁,” (3) is reduced to (4):

𝑉 = ^{𝑃𝛽−𝑃𝛽𝑂𝑁}

𝑃_{𝛽𝑂𝐹𝐹}−𝑃_𝛽𝑂𝑁𝑉_𝑂𝐹𝐹 (4)

This adaptive rule is applied only for values of 𝑃_𝛽𝑂𝑁 < 𝑃_𝛽 < 𝑃_{𝛽𝑂𝐹𝐹}, for 𝑃_𝛽 > 𝑃_{𝛽𝑂𝐹𝐹} 𝑡ℎ𝑎𝑛 𝑉 = 𝑉_𝑂𝐹𝐹, 𝑃_𝛽 < 𝑃_𝛽𝑂𝑁 𝑡ℎ𝑎𝑛 𝑉 = 0.

In the aDBS mode, the device was therefore programmed to deliver a 130 Hz stimulation with a 60 µs pulse and an amplitude that linearly changed, according to beta power recorded, between 0 V and the effective stimulation amplitude calculated for each patient. Conversely, in the cDBS mode, the device delivered a 130 Hz stimulation with a 60 µs pulse at the effective stimulation amplitude.

Thus, in the aDBS mode, the stimulation amplitude could not exceed the effective stimulation amplitude delivered continuously in the cDBS mode. The prototype recorded the stimulation amplitude delivered during the aDBS session.

37 3.3.1.4 Technical and clinical outcomes

The technical outcomes encompass the functioning of the aDBS device and its energy efficiency.

The former included the assessment of the efficacy of DBS stimulation alone, in both modalities (aDBS and cDBS). To make this assessment, we calculated the UPDRS III improvement in the stimON/medOFF condition with the stimOFF/medOFF condition. It also included the assessment of its ability of the aDBS device algorithm to follow beta changes by comparing the stimulation

amplitude delivered during OFF and ON Med.

The energy efficiency of the prototype in the aDBS and cDBS modes was evaluated using the total electrical energy delivered (TEED) per unit (Koss et al., 2011) with a reference impedance of 0.5 kΩ.

The clinical outcomes were the aDBS clinical effect compared to cDBS evaluated through the UPDRS III and the UDysRS scores when the patient had both DBS and medication

(stimON/medON condition).

3.3.1.5 Statistical analysis

We first verified that clinical scale scores (UPDRS III and the Unified Dyskinesia Rating Scale [UDysRS III-IV]) were normally distributed through the single-sample Shapiro-Wilk test (p > 0.05); therefore parametric statistical analyses were performed.

We then calculated the stimulation-induced improvements for the UPDRS III clinical scale in the stimON/medON condition compared to the baseline condition (stimOFF/medOFF), and we compared the UDysRS scores during cDBS and aDBS in stimON/medON. We used a repeated measures one-way ANOVA for both the UPDRS III and the UDysRS scores.

To verify the effectiveness of the device delivering dual-mode stimulation (aDBS and cDBS), we compared the UPDRS III at the baseline and at the stimON/medOFF condition (one-way ANOVA).

As an adjunctive analysis on the effect of DBS alone (stimON/medOFF condition), we examined the change in UPDRS III items on the side contralateral to stimulation.

Then, to study the other secondary outcomes (functioning of the aDBS algorithm), we used a repeated measures one-way ANOVA to compare the mean amplitude delivered by the device during the OFF and ON Med periods.

We used a repeated measures one-way ANOVA to compare TEED in the aDBS and cDBS modes.

Power saving was calculated as the mean percentage of change in TEED in the aDBS compared to the cDBS modes.

38

Statistical analyses were performed using the Matlab software (The MathWorks, Natick, USA).

Differences were considered significant at p < 0.05. Throughout the text, values are given as mean

± standard deviation (SD).

3.4 Results

We found that 14 patients out of 17 met the neurophysiological inclusion criterion displaying beta activity. Fourteen PD patients (nine men and five women; age 56.4 ± 8.4 years, disease duration 12.6 ± 5.3 years) were enrolled. Of them, four withdrew from the study: three patients left the study during the second experimental session because of intolerability to medication withdrawal, and Patient 3 left the study during the third experimental session because of fatigue. Ten patients completed all of the experimental sessions: four patients first received aDBS and six patients first received cDBS. Consistent with recent findings (Quinn et al., 2015), we observed that the 100% of AR patients and the 62.5% of TD patients showed beta band oscillations. Moreover, women are more likely to show beta activity than men (100% vs 70%), and no significant differences were found for age and disease duration.

3.4.1 Technical results

The stimulation voltage changed according to the changes in beta power. The beta band power decreased when patients completed the transition from the med OFF to the med ON state.

Accordingly, the stimulation amplitude linearly decreased from the med OFF to the med ON state (Fig. 3.3b). The stimulation amplitude significantly decreased from the med OFF to the med ON conditions in all cases (med OFF vs med ON: 2.1 ± 1.2 V vs 0.8 ± 0.9 V; p = 0.007). The mean TEED with aDBS (44.6 ± 47.9 µW) was significantly less than with cDBS (158.7 ± 69.7 µW;

p = 0.0005) (Fig. 3.3) with an average power saving value of 73.6 ± 22.9% in aDBS compared with cDBS.

39

Figure 3.3 (a) Stimulation amplitude course in one patient, (b) total electrical energy delivered (TEED) comparison

3.4.2 Clinical results

We first assessed the ability of the device to deliver an effective stimulation during both aDBS and cDBS without levodopa, according to the decrease of UPDRS III score: the UPDRS III score significantly decreased from the stimOFF/medOFF condition to stimON/medOFF condition during cDBS (36.6 ± 16.2 vs 30.8 ± 14.3; p = 0.03) and during aDBS (37 ± 16.8 vs 33.1 ± 16.4; p = 0.002) sessions (Fig. 3.4).

The clinical scores were not significantly different between the two treatment days at the baseline (stimOFF/medOFF) (UPDRS III: aDBS vs cDBS: 37 ± 16.8 vs 36.6 ± 16.2; p > 0.05).

When the patient was under the effect of both levodopa and DBS, we observed a similar improvement in global motor symptoms due to aDBS and cDBS combined with levodopa, but aDBS had a remarkable lowering effect on dyskinesias compared to cDBS. In fact, comparing the stimulation-induced effect, the one-way ANOVA showed that aDBS and cDBS induced similar improvements in UPDRS III score when combined with levodopa in stimON/medON conditions (UPDRS III, aDBS vs cDBS: −46.1 ± 10.5% vs −40.1 ± 17.5%; p > 0.05, Fig. 3.5a).

Conversely, aDBS and cDBS differentially modulated in terms of the UDysRS score. During the aDBS session, patients experienced significantly less dyskinesias, compared with in the cDBS session, in the stimON/medON conditions (aDBS vs cDBS: 11.6 ± 69 vs 15.0 ± 8.7; p = 0.01, Fig.

3.5b)

40

Figure 3.4 UPDRS III score under conventional deep brain stimulation (cDBS) and adaptive DBS (aDBS)

Figure 3.5. (a) UPDRS III score percentage improvement, adaptive deep brain stimulation (aDBS) vs conventional DBS (cDBS); (b) UDysRS score improvement, aDBS vs cDBS.

41

4 Exploring local field potential-based adaptive deep brain stimulation feasibility: Eight hours of

monitoring

4.1 Rationale for the study and objectives

The proof of concept study reported in the previous chapter showed that under defined and controlled circumstances, adaptive deep brain stimulation (aDBS) is as effective as cDBS and avoids unwanted side effects such as dyskinesias. The same results may be not valid during a longer period of time and during a common daily medication administration (not increased of the 50%). In particular, to our knowledge, no one has explored the correlation of beta oscillations with the patient clinical state during daily activities and for a long period of time. All the studies that investigated LFP response to levodopa have been restricted to a single administration (Priori et al., 2004; Ray et al., 2008; Kuhn et al., 2009). A lack of correlation would prevent the feasibility of LFP-based aDBS. We therefore designed a two-day experimental session. During the first day, we monitored the beta band power changes and the clinical state of the patient; during the second day, we also applied aDBS to test whether or not it was possible to follow clinical fluctuations, to explore clinical results and to propose future steps to improve the methodology.

4.2 Methods

4.2.1 Patients’ inclusion criteria

All patients effected by PD who had undergone to DBS electrode implant in the STN with respect to the surgery eligibility criteria (L.I.M.P.E.) not showing complications after surgery were included as candidates for the study. The experimental protocol took place after the surgery for electrode placement and the surgery for the implant of the pulse generator. Deep brain stimulation electrodes were externalized to allow for stimulation and recording. Patients not showing significant LFP activity in the beta band were excluded (see Section 4.2.3). Each patient underwent to two

experimental sessions the fifth and the sixth day after the surgery for the electrode implant to ensure impedance stability (Rosa et al. 2010). Each experimental session lasted seven to eight hours and began after 12 hr of medication withdrawal.

42

4.2.2 Experimental protocol

The experimental protocol was composed of two sessions (i.e. two days) (Fig. 4.1).

Figure 4.1 Experimental protocol First session

Local field potential recordings at rest, contact selection, and band definition. We recorded the LFPs at rest from both the hemispheres in bipolar configuration from the contact pairs (0-1,0-3, 1-3) of the externalized DBS electrodes (Model 3389, Medtronic). An Ag-AgCl surface electrode was positioned on the right shoulder as a reference. The LFPs were recorded with an instrumental amplifier (Model Grass ICP511, gain 80 dB, passband 3–30 Hz, notch ON) and digitalized (CED Micro-1401, Cambridge Design, UK). The LFPs were sampled at 256 Hz and their PSDs were visualized online by means of the software Spike 2 (Spike 2, Cambridge Design, UK), using a Hanning window of 1 s averaged over 50 s to 60 s. Based on intra-surgery stimulation tests we choose the contact pair allowing for stimulation at an effective contact and showing a detectable beta peak. The band was selected at +/−2 Hz around the peak frequency (Ray et al., 2008; Kuhn et al., 2009). Electrode impedances were acquired at 30 Hz using an impedance meter (Model EZM 4;

Grass, USA).

Neurophysiological monitoring. The external portable device (see Chapter 2) was connected to the selected contact and shoulder reference to record LFPs during the day for up to eight hours. The device stored in not volatile memory a value of the power of the selected band every 2 s. In addition, the 2.4 GHz RF interface was programmed to send the module of the FFT between 5 Hz and 30 Hz every second. The RF link on the device board covered a short range, allowing us to track the LFP data during daily activity occurring in the hospital room and not outside. The device

43

was connected to the belt and allocated in a pouch, thus allowing the patient to carry out daily activities. A wearable accelerometer was placed on the arm contralateral to the stimulation to collect kinematic data. The LFPs and acceleration data were stored in real time on the PC by means of a custom script written in Matlab for reading data arriving at a USB access point (Fig. 4.2).

Figure 4.2 Data collection

Pharmacological treatment and clinical monitoring. The clinicians subministered to the patient a levodopa therapy similar to the regular treatment the patient had before the operation, but decreased the dose to accomplish the state of the patient improved by the “stun” effect of the surgery for the DBS electrode implant. If the patient was used to taking dopamine agonists, these were substituted with levodopa therapy when possible. In cases that such a substitution was not possible, the patient was included in the study anyway under the hypothesis that the dopamine agonist has the same effect as levodopa on LFPs spectral content (Priori et al., 2004). The clinical state of the patients were monitored by an unblinded neurologist who annotated on a clinical diary the time and the dosage of the levodopa therapy and the state of the patient. The clinical state of the patients were evaluated in the OFF condition, during the peak dose (Med ON) and at the end of the dose before the successive pharmacological assumption (Med OFF), with constant delays when possible. For each evaluation, the neurologist gave scores on the following rating scales: UPDRS III and UDysRS. At least two pharmacological cycles for each patient have been monitored (Fig. 4.1). A clinical diary was also filled by the patient every 30 min as qualitative support in the clinical monitoring process, but its contents do not appear in the following the analysis.

44 Second session

Tuning the stimulation threshold. With the use of our external device, the stimulation voltage was increased in steps of 0.5 V until the appearance of side effects. At each step, a neurologist evaluated the contralateral motor improvement. The voltage threshold was chosen as the value which

provided the greatest clinical effects without exacerbating side effects.

Setting and Testing aDBS. The aDBS device was programmed as described in Section 3.3. The stimulation voltage changed linearly between the threshold voltage value and zero. The beta band analysis was performed in real time as described in Chapter 2, and the stimulation voltage was adapted proportionally on the basis of the power range given by the value of the beta power in Med OFF condition (at rest) and the value in Med ON condition.

Neurophysiological and clinical monitoring. The neurophysiological and clinical monitoring were done following the same methods used during the first session.

Results obtained during the two days of the experimental session were excluded if the offline analysis of rest LFPs failed to show a significant oscillation.

4.2.3 Signal processing and statistical analysis

Local field potentials analysis

Local field potentials of nine patients were recorded from the 18 STNs five days after the implant of DBS electrodes. The patients were 12 hr without medication at the time of the recordings. The LFPs were recorded through the Grass amplifier (bandwidth 1–100 Hz, gain 80 db, notch ON) and

sampled at 256 Hz. Signals were filtered offline between 2 and 50 Hz in forward and backward directions (“filtfilt” Matlab) to avoid any phase distortions. The minimum common length of the data for all patients was 50 s. To separate the background noise to the oscillation component, we applied the coarse-grained FFT analysis as proposed in (He et al., 2010; He et al., 2014). This method centers on the hypothesis that the background noise is a “scale free” process having a power spectrum following the rule 1/𝑓^𝛼. Under this assumption, the background neural noise is not given by the averaging of transient, non-stationary oscillations, but is the result of arrhythmic activity on a fractal domain.

The coarse-grained FFT analysis, therefore, provides three spectral estimations, Praw (power spectrum of the raw data), Posc (power spectrum of the oscillatory), and Pfractal (power spectrum

45

of the "scale-free" background noise). A peak in the Praw component was considered independent of the background noise when a separation of the confidence interval (95%) occurred.

To test the independence between impedances of the recording electrodes and the amplitude of the alpha or beta peak we calculated the Pearson coefficient.

Local field potentials and clinical state correlation

Only the neurophysiological data of patients showing a significant beta oscillation were included in this study. For these patients, the band power time course was analyzed offline. The band power was interpolated to have a value every second and then smoothed at an average of 10 min average.

In correspondence of the clinical evaluations, the mean value of the band power was extracted. The absolute values of the band power and the percentage change with respect to the OFF Med state (band power OFF Med – band power current state / band power OFF Med state) were then linearly correlated to the absolute UPDRS III percentage changes from the OFF Med state (UPDRSIII OFF Med – UPDRSIII OFF current state/ UPDRSIII OFF OFF Med state) (here OFF Med refers to the first OFF Med state obtained after 12 hr of medication withdrawal). We tested the normality distribution of the data with the single sample Shapiro-Will test (p < 0.05), and we then applied parametrical statistics.

Clinical data

We collected the total UPDRS III and the UDysRS (part 3 and 4) five times during the whole experimental session: once at baseline, twice when the patient was in peak dose (Med ON), and twice when the effect of medication ended (Med OFF). We used these data to compare the clinical outcomes in the two days, and establish whether aDBS was effective.

We used the percentage value at each time point for further analyses (V%t). We verified

distribution normality through the single-sample Shapiro-Wilk test (p>0.05), to allow the use of parametric statistics.

Three main factors could influence the values of clinical data: the day of the session (Day 1, no aDBS; Day 2, with aDBS); the medication condition (Peak dose, Med ON vs End dose, Med OFF);

and the medication session (session 1, first morning dose; session 2, second daily dose). For this reason, we first run a three-way Analysis of Variance (ANOVA) with factors “therapy” (2 levels, Day 1 and Day 2), “condition” (2 levels, Med ON and Med OFF), and “session” (2 levels, Session 1 and Session 2). When the three-way ANOVA showed a significant effect (p<0.05) for the factors

and the medication session (session 1, first morning dose; session 2, second daily dose). For this reason, we first run a three-way Analysis of Variance (ANOVA) with factors “therapy” (2 levels, Day 1 and Day 2), “condition” (2 levels, Med ON and Med OFF), and “session” (2 levels, Session 1 and Session 2). When the three-way ANOVA showed a significant effect (p<0.05) for the factors